Traffic sensors provide real-time measurements of current traffic conditions, whereas traffic management applications require forecasts of future traffic speed or flow. This study considers the incremental expansion of an operating sensor network, in which a pre-trained network-level forecasting model must predict traffic states at sensor locations that were absent during training. Many spatio-temporal graph neural networks rely on sensor-specific embeddings and graph connections learned from data-rich training networks. These representations are undefined for previously unseen locations, limiting the direct application of pre-trained adaptive-graph forecasters during sensor-network expansion. To address this cold-start problem, we propose support-conditioned sensor-adaptive meta-graph learning (SC-SAMG), which derives target-node representations and spatial dependencies from a short support period. The framework combines a support-set encoder, a task-specific graph learner, and first-order meta-learning to adapt the network-level forecaster using one to seven days of target observations. Experiments on four traffic benchmarks evaluate forecasts of speed or flow over the next 15–60 min under a leakage-controlled held-out-node protocol. SC-SAMG consistently outperforms fine-tuned gated recurrent unit (GRU), adaptive graph convolutional recurrent network (AGCRN), and diffusion convolutional recurrent neural network (DCRNN) baselines. It reduces mean absolute error (MAE) by up to 11% relative to the adaptive-graph baseline and by up to 7% relative to the diffusion convolutional baseline. These results demonstrate the potential of support-conditioned graph adaptation for incorporating previously unseen sensor locations into existing network-level traffic forecasting systems.
Can Wang, Zhiyu Wang, Weijie Wang et al.· Italian National Conference...· 0 citations
Road traffic accident consequences are multidimensional, involving fatalities, injuries, and property loss. Existing studies have mainly focused on single outcomes, limiting the understanding of heterogeneous mechanisms across different consequence dimensions. Based on road accident data from Yancheng City in 2022, this study develops an integrated framework combining multi-output prediction, NSGA-II multi-objective optimization, SHAP-based interpretation, LOWESS nonlinear analysis, and DirectLiNGAM causal inference. The results show that the optimized Voting ensemble achieved competitive and comparatively balanced performance across the three accident-consequence dimensions. Road category, traffic control type, junction/road-segment type, and crash-cause category are identified as key influencing factors, with differentiated effects across accident consequences. POI variables exhibit nonlinear and threshold effects, while causal analysis further indicates that road infrastructure and traffic control conditions are positioned upstream in the formation mechanism of accident consequences. This study provides evidence for multidimensional accident-consequence category prediction and differentiated traffic safety management, rather than traditional continuous regression-based modeling.
Yanni Ju, Wanqiu Li, Di Tang et al.· Applied Sciences· 0 citations
This study proposes an enhanced key-node-driven framework for traffic flow prediction in large-scale transportation networks. Building upon the classical K-shell decomposition, the proposed method integrates traffic-flow-based weighting to jointly capture structural hierarchy, functional relevance, and global topological influence of network nodes. By ranking node importance through composite indicators lambda-c, lambda-f, and lambda-s, the framework identifies structurally dominant nodes and evaluates their effectiveness across multiple traffic prediction models. Comprehensive experiments conducted on three benchmark datasets—PEMS04, METR-LA, and PEMS-BAY—demonstrate that graph-based spatiotemporal models such as ST-GCN, GraphWaveNet, DCRNN, STDN, STTN, and SWAVE maintain high predictive accuracy even under reduced node coverage. In particular, DCRNN exhibits strong robustness in capturing dynamic spatiotemporal dependencies, while STTN effectively models long-term temporal patterns. Overall, the results highlight the robustness and versatility of the proposed framework in real-world traffic scenarios, demonstrating its potential to enhance prediction accuracy and scalability through key-node identification and selective coverage analysis.
Jing Gan, Dongmei Yan, Yue Wang et al.· Systems· 0 citations